DataSpeaks Interactions® Software –
A Killer App for 21st Century Systems Science and P4 Medicine?
CURTIS A. BAGNE, PH.D.
DATASPEAKS, INC.
International Symposium, Institute for Systems Biology, April 2008
1 Background
Dr. Leroy Hood has represented systems biology as shown:
DataSpeaks Interactions® Software embodies a
computational algorithm that measures “interactions” or
“edges” as interactions over time when the “elements” or
“nodes” are action variables. Furthermore, DataSpeaks
Interactions® measures coordination of action – the
harmonious functioning of parts for effective results – as an
“Emergent” system property.
Scientists and engineers have developed great tools to identify
and measure parts of cells, organs, organisms, ecosystems and
their environments. Now DataSpeaks Interactions® can help users
describe and predict how such parts interact to form systems that
work:
• Function internally (Demonstration 1),
• Respond to their environments including treatments
(Demonstration 2), and
• Act as agents on their environments.
New measures of interaction over time will help users build
systems science on more reductionistic approaches to basic and
applied science. DataSpeaks Interactions® can help take us from
“Genomes to Life.” As such, DataSpeaks Interactions® deserves
to be evaluated as a tool and potential killer app that can help
expedite 21st Century science and P4 medicine.
2 Technical Overview
DataSpeaks Interactions® applies to time series
data with at least two but preferably many
more repeated “measurements” to measure
the amount, strength and positive or negative
direction of evidence for interactions over time
between at least one action variable functioning
as an independent variable and at least one
action variable functioning as a dependent
variable.
systems include levels of nutrients, allergens,
pollutants and other stimuli as well as doses
of drugs.
Action variables are variables that can change
and fluctuate in level over time for individual
systems.
Most laboratory variables in medicine and
measures of vital signs, symptoms, mental and
motor performance, and quality of life can
be investigated as dependent action variables.
Although death rates can be investigated as
action variables for whole populations, vital
status and survival time are not action variables
for individual patients. Electrophysiological
measures are action variables.
Independent action variables include levels,
concentrations and doses of “elements” that
form “nodes.” Endogenous independent
action variables include levels of gene
expression and concentrations of proteins,
metabolites and other biological constituents.
Exogenous or environmental independent
action variables that sustain and can perturb
Genome and genotype are categorical or
classification variables – not action variables
– for individual organisms. DataSpeaks
Interactions® does not apply to classification
variables. However, DataSpeaks Interactions®
can help elucidate what genetic differences
mean in terms of how different organisms
work and how genetically similar organisms
(e.g., man, mouse) can work quite differently.
DataSpeaks Interactions® can help users
transcend the limits of data snapshots, which
include change scores, to the advantages of
multiple time series data or data movies.
Compared to data snapshots, data movies
can account for time, time dependent
system dynamics and can include orders of
magnitude more information to understand
how individual systems work. Furthermore,
data movies might be the best way to
investigate coordination of action as a time
dependent “emergent” system property. Many
data collection technologies already collect or
are capable of collecting data movies.
Demonstrations 1 and 2 illustrate
fundamental innovations in scientific
methodology using some of the most
rudimentary features and capabilities of
DataSpeaks Interactions® Software.
DataSpeaks Interactions® Software –
A Killer App for 21st Century Systems Science and P4 Medicine?
CURTIS A. BAGNE, PH.D.
DATASPEAKS, INC.
International Symposium, Institute for Systems Biology, April 2008
3 Demonstration 1
Measuring Coordinated Biological Function as
a Time Dependent Emergent System Property
The time series in Figure 1 are for two “elements” or “nodes.”
These data were collected every five minutes for about 12
hours from hypophyseal portal blood from one ewe. Our task
is to demonstrate the capabilities of DataSpeaks Interactions®
and help validate this software by measuring a well known
interaction over time between two hormones.
Figure 1. Data for Demonstration 1
10,000
Luteinizing hormone (LH), ng/mL
1000
100
Figure 1 is a data movie with two pixels or time series
variables and 143 repeated measurements (movie frames) for
one individual. See the coordination of action. Now we will
measure this coordination as interactions over time so that the
coordinated action can be investigated more scientifically.
10
1
Gonadotropin releasing hormone (GnRH), pg/mL
0.1
0
100
300
200
400
600
500
700
Time, Minutes
3.1 Developing “Data Movies,” Distribution of
Potential Interaction Scores and Results
DataSpeaks Interactions® was used to “develop” the data movie in Figure
1 by measuring both pairwise directional interactions (GnRH to LH, LH
to GnRH) with a user specified scoring protocol that yielded a detailed
but easy to summarize 4-dimensional array of 1,008 (6 x 6 x 7 x 4)
standardized interaction scores that accounted for:
• Level of the independent variable (6 levels of dimensional
resolution),
• Level of the dependent variable (6 levels of dimensional resolution),
• Any time delay of apparent effect of independent events on
dependent events (7 levels, 0 thru 6), and
• Any time persistence of effects of independent events on dependent
events (4 levels, 1 thru 4).
This array of 1,008 standardized interaction scores was summarized by
selecting the most extreme positive or negative value, which is 76.028. On
30 of 32 repeated measurement times in Figure 1 when a specific type of
independent GnRH event was present, a specific type of dependent LH
event also was present. In addition, the same specific type of dependent
LH event never was present on any of 111 times when the same specific
type of independent GnRH event was not present.
This summary interaction score (76.028) is one score from a distribution
of potential scores, defined by the data in Figure 1 in combination with
the scoring protocol. This distribution, shown in Figure 2, has a mean of
zero and a standard deviation of 1. The summary score of 76.028 indicates
substantial evidence for a GnRH to LH interaction. This confirms what
most people are apt to think after looking at Figure 1. However, unlike
subjective impressions, the software yields a reproducible quantitative
result and applies to problems beyond unaided human capabilities.
Figure 2
Probability
0.20
0.18
0.16
0.14
0.12
0.10
0.08
0.06
0.04
0.02
0
-8 -6 -4 -2 0 2 4 6 8 10
Observed score = 76.028
p = 7.61597E-029
20
30
40
50
60
70
80
DataSpeaks Interactions® Software –
A Killer App for 21st Century Systems Science and P4 Medicine?
CURTIS A. BAGNE, PH.D.
DATASPEAKS, INC.
International Symposium, Institute for Systems Biology, April 2008
Figure 3
80
Strength of Evidence
1.0
Observed summary score
70
50
0.6
40
0.4
30
Amount of Evidence
20
Strength of Evidence
0.8
60
Amount of Evidence
Figure 3 illustrates performance characteristics
of two different and complementary types of
measures that quantify:
• Amount and direction of evidence versus,
• Strength and direction of evidence for
the interaction over time for the GnRH
(independent variable) to LH (dependent
variable) interaction.
These interaction scores were computed
iteratively across the 143 repeated measurement
times (715 minutes) that are shown in Figure 1.
0.2
10
0
0.0
0
100
300
200
400
600
500
700
Time, Minutes
Figure 4
Figure 4 summarizes the 4-dimensional arrays of
1,008 standardized interaction scores as functions
of their four dimensions or analysis parameters.
Summary scores are the most extreme positive or
negative values in a row, a column or an overall array.
Part A: GnRH Level
Interaction Score
80
70
70
Observed
summary
score
60
50
Parts A and B of Figure 4 summarize the GnRH
to LH interaction as functions of the levels of the
interactants expressed as residuals from a linear
regression line that was used to de-trend the time
series before measuring the interactions over
time. Taken together, Parts A and B show that at
least moderate level LH events are most apt to be
associated with high level GnRH events. Temporal
asymmetries in Parts C and D suggest that high
levels of GnRH are more predictive of high levels
of LH than vice versa, especially at time delay level
1. More generally, DataSpeaks Interactions® can be
used to explore the temporal criterion of causal
relationships.
Part B: LH Level
Interaction Score
80
40
GnRH to LH
60
50
40
30
30
20
20
10
10
0
-25 -20 -15 -10 -5
0
5
10 15 20 25
GnRH to LH
0
-400 -300 -200 -100
GnRH Level
60
Interaction Score
100
80
Observed
summary score
GnRH to LH
60
40
40
20
LH to GnRH
0
20
0
-20
GnRH to LH
-40
0
(0)
100 200 300 400
Part D: Time Persistence
Part C: Time Delay
Interaction Score
100
80
0
LH Level
LH to GnRH
-20
-40
1
(5)
2
(10)
3
(15)
4
(20)
5
(25)
6
(30)
1
(0)
2
(5)
3
(10)
4
(15)
Time Delay
(minutes)
Time Persistence
(minutes)
• Enables both global and
detailed assessments of
coordinated system function;
• Has high potential value for:
• Elucidating and visualizing
normal coordination,
• Objective diagnoses of
disorders of coordinated
function,
• Elucidating mechanisms
of action,
• Functional biomarkers
and phenotypes.
3.2 Prospectus – Demonstration 1
Measurement of
coordination of action
as interactions over
time as illustrated by
Demonstration 1:
• Can be extended to
hundreds or thousands of
time series variables for highthroughput investigations
as with functional brain
imaging, which is nondestructive;
DataSpeaks Interactions® Software –
A Killer App for 21st Century Systems Science and P4 Medicine?
CURTIS A. BAGNE, PH.D.
DATASPEAKS, INC.
International Symposium, Institute for Systems Biology, April 2008
4 Demonstration 2
A Placebo Controlled Single Group Randomized Controlled Trial
(Clue – Randomize doses to time periods.)
“FDA’s evaluation methods have remained
largely unchanged over the last half century.”
(FDA Science Board, FDA Science and Mission at
Risk, November 2007)
Randomized controlled trials (RCT) are the
best way to assess causality. However, current
first generation RCT designs, which provide
much of the evidentiary basis of the current
public health (not personalized or P4) approach
to medicine, predate:
• Modern computers,
• The genomic revolution and systems biology,
• High-throughput and Internet enabled
measurement technologies and
• Major companies such as Microsoft and
Google, “the two leading candidates for
Web supremacy,” that now seek to provide
software-based services to improve health
(“Dr. Google and Dr. Microsoft: Two
Giants Have Plans to Change Health Care,”
New York Times, 8/14/07).
4.1 Confounding in Current First Generation RCT Designs
Not only have RCT designs “remained largely
unchanged over the last half century,” but
such designs also confound.
Current first generation RCT designs that:
• Randomize patients to different groups to
assess causality (1) confound individuality
with measurement error and (2)
confound true responders to active
treatment with responders on active
treatment that would have responded to
placebo,
• Randomize patients to different groups
defined by different doses of a particular
type of treatment in essentially the same
way that patients are randomized to groups
defined by different types of treatment (3)
confound dose with type of treatment
as in current first generation placebo
controlled RCT designs and
• Perform statistical tests on health variables
or changes in health variables (4) confound
treatment effects and how treatment
effects are valued.
We can end these four types of confounding
to help:
• Revitalize the pharmaceutical industry by:
• Translating advancements in life sciences
into better products and health,
• Improving productivity and research
efficiency and
• Reducing safety problems;
• Improve research investment returns;
• Save lives, avoid harm and avoid legal
liability;
• Avoid loss of products as with Vioxx
(Merck, pain) and Bextra (Pfizer,
pain) and loss of potential products
as with torcetrapib (Pfizer, cholesterol
management);
• Improve the economic competitiveness of
states and nations;
• Improve public credibility and support for
science.
Table 1 illustrates a second generation RCT
design that does not commit these four types
of confounding. Such new designs can be
applied to evaluate drug treatments for chronic
health problems of function such as chronic
pain, hypertension, lipid disorders, diabetes,
endocrine disorders and neuropsychiatric
disorders.
Second generation RCT designs are possible
when both independent and dependent
variables can be investigated as action variables
– variables that vary and fluctuate in level over
time for individuals. Table 1 uses mock data for
treatment and health.
Week
Table 1
1
Patient
Variable
Patient 1
Dose
DBP3
ED4
Energy
Patient 2
Dose
DBP
ED
Energy
Patient 3
Dose
DBP
ED
Energy
2 3
Pair 1
4
5
6 7
Pair 2
Period
1
8
9
10 11 12 13 14 15 16
Pair 3
Pair 4
Period
2
1
Period
2
1
Period
2
1
Overall
Interaction
score
Direction1
2
20 20 40 40 0 0 40 40 40 40 20 20 80 80 40 40
96 98 85 81 91 96 84 87 80 78 93 98 82 77 81 78
3 2 2 3 2 1 2 1 3 2 1 2 3 2 2 2
4 3 4 5 4 3 4 3 2 4 3 3 4 4 2 3
-8.92
0.74
1.46
0 0 20 20 40 40 0 0 20 20 80 80 80 80 40 40
98 91 89 88 89 84 96 98 89 93 86 80 76 75 80 92
Harm
Score
Weight2
Harm
Score
+
8.92
-0.74
1.46
4
2
2
4.64
-8.56
-
8.56
4
2
3
5.93
3.05
+
+
5.93
3.05
1
2
40 40 20 20 0 0 80 80 20 20 40 40 20 20 80 80
76 79 74 80 88 90 78 68 75 77 81 78 73 79 76 82
1 3 2 2 1 1 2 4 2 2 2 1 3 2 3 5
4 3 2 2 3 4 3 2 5 3 4 3 4 4 1 2
-7.81
5.04
-2.67
+
7.81
4
-5.04
1
-2.67
1
Group Average
2
2
1
3
2
2
2
2
3
3
2
1
1
2
2
3
2
2
3
2
2
3
4
3
4
4
3
3
4
2
1. Direction indicates whether higher values on a health variable were considered to be beneficial or harmful for each particular patient.
2. Weights are used to specify the relative value of treatment effects with respect to different health variables for each particular patient.
3. DBP = Diastolic Blood Pressure
4. ED = Erectile Dysfunction
6.61
3.92
5.06
t=6.29
p=.0242
DataSpeaks Interactions® Software –
A Killer App for 21st Century Systems Science and P4 Medicine?
CURTIS A. BAGNE, PH.D.
DATASPEAKS, INC.
International Symposium, Institute for Systems Biology, April 2008
Each of the nine health variable specific benefit/harm scores in
Table 1 is one score from a distribution of potential scores that has
a mean of zero and a standard deviation of 1. Figure 5 shows all
nine of these distributions. Standardization helps make it feasible
to compare and combine observed benefit/harm scores from
different distributions.
Figure 5: Distributions of potential
benefit/harm scores for Demonstration 2
Probability
0.8
Patient 1
0.7
0.6
Figure 6: Benefit/harm as a function
of drug dose for Demonstration 2
Part A: Patient 1
Benefit/harm score
10
Part B: Patient 2
Benefit/harm score
10
8.92
8
8
6
6
4
4
2
2
0
0
-2
-2
-4
-4
-6
0.5
-6
0
0.4
20
40
0.1
-4
-2
Probability
0.8
0
2
Benefit/harm score
4
6
8
10
Patient 2
DBP
ED
Energy
0.7
0.6
Benefit/harm score
10
8
8
6
6
4
4
2
2
0
-2
-2
-4
-4
0.2
-6
Observed score
-8
-6
-4
-2
Probability
0.8
0
2
Benefit/harm score
-6
0
Observed score
4
6
8
20
10
60
80
0
20
60
80
Figure 7:
Monitoring evidence over time for apparent
benefit/harm for Demonstration 2
0.5
0.4
0.3
Part A: Patient 1
Observed score
Benefit/harm score
10
Observed score
Observed score
0
-8
40
Dose
Patient 3
0.6
0.1
40
Dose
0.7
0.2
80
Part D: Group Average
Benefit/harm score
10
0.4
0.3
60
Energy
Overall B/H
0
0
40
Dose
0.5
0.1
20
Part C: Patient 3
0
-6
0
DBP
ED
Observed score
Observed score = 8.92
0.2
-8
80
Dose
Observed score
0.3
60
-6
-4
-2
0
2
Benefit/harm score
4
6
8
10
Figure 6 shows health variable specific and overall benefit/harm as
a function of dose for each patient. The optimal minimal doses for
patients 1, 2 and 3 were 40, 80 and 20 respectively. Part D shows
the group averages. Current first generation RCT designs do not
identify optimal minimal doses for individual patients or for groups.
Figure 7 shows how evidence for benefit/harm can be monitored
over time (repeated measurement times) for each individual patient
and for the group on average. Evidence from such monitoring
can help assure that harm does not exceed benefit for any patient.
Individual patients in RCTs that evaluate drugs for chronic health
problems can be treated as patients – not research subjects. Research
and clinical practice can be more integrated.
Second generation RCT designs as illustrated by
Demonstration 2 are more scientific, informative, efficient
and ethical than current first generation RCT designs.
Part B: Patient 2
Benefit/harm score
10
8.92
8
8
6
6
4
4
2
2
0
0
-2
-2
-4
-4
-6
-6
0
2
4
6
8
10
12
14
16
0
2
4
6
Week
8
10
12
14
16
14
16
Week
DBP
ED
Energy
Overall B/H
Part D: Group Average
Part C: Patient 3
Benefit/harm score
10
Benefit/harm score
10
8
8
6
6
4
4
2
2
0
0
-2
-2
-4
-4
-6
-6
0
2
4
6
8
Week
10
12
14
16
0
2
4
6
8
10
12
Week
DataSpeaks Interactions® Software –
A Killer App for 21st Century Systems Science and P4 Medicine?
CURTIS A. BAGNE, PH.D.
DATASPEAKS, INC.
International Symposium, Institute for Systems Biology, April 2008
4.2 DataSpeaks Interactions® enables fundamental innovations in RCT design:
• Causality can be assessed by randomizing
various doses to different periods of time
and measuring interactions over time for each
individual before statistical analyses and
hypothesis testing for group data.
• Dose can be investigated as an
independent action variable for each
patient. Patients were not classified into
groups defined by doses.
• No patient was randomized to placebo
only, only an insufficient or ineffective
dose, or only an excessive or harmful
dose. This is an important ethical issue.
• True response to active treatment can
be differentiated from placebo response
for each individual patient.
• Optimal minimal doses can be ascertained
for each individual as illustrated by Figure 6.
• Commensurability. Beneficial treatment
effects (effectiveness) and harmful treatment
effects (safety) with respect to various health
action variables can be measured and used to
compute one overall benefit/harm score for
each patient using explicit directionalities
(positive or negative) and weights (indicating
clinical and personal significance) as shown
in Table 1.
• Safety and effectiveness can be profiled
across health variables and balanced
scientifically.
• One hypothesis and one statistical test
in one RCT can evaluate benefit/harm
with respect to many health action
variables.
• Treatment evaluations can be
comprehensive of effects on all health
action variables that can be measured
repeatedly.
• Although hypotheses do need to
be tested with predetermined
directionalities and weights, ex post
facto exploratory investigations can be
conducted with different directionalities
and weights.
• One way to transform interaction
scores into benefit/harm scores is to
investigate how and to what extent
interaction scores predict survival times
and death rates.
• The null hypothesis of no overall benefit/
harm was rejected in the positive or
beneficial direction using a two-tailed single
group t-test on the mean. No statistical test
was performed on health variables or changes
in health variables.
• Time delays and persistencies in treatment
response can be accounted for as indicated
by Figure 4, Part C and Part D for
Demonstration 1.
• Data from 16 repeated measurements
for each patient were used to overcome
measurement error, increase statistical
power and help achieve statistical
significance with only three patients (see
Table 1, page 4).
• All of the above offer promise to help
investigators identify genetic and other
predictors of differential responses to
treatments and to make genotyping more
valuable.
• Potential new gold standards for clinical
research and practice can be integrated.
The market for DataSpeaks Interactions® in
clinical practice is larger than its market as a
research tool.
Comparative head-to-head RCTs can be
conducted by randomizing patients to
different types of treatment before conducting
a single group RCT for each type of
treatment.
5 DataSpeaks Interactions® Can Help Expedite 21st Century Systems Science
Measurement of coordinated action as a
time dependent “emergent” system property
(Demonstration 1) and measurement of the
benefit/harm of treatments (Demonstration
2) makes such concepts amenable to scientific
investigations. New measures have a history of
advancing science and commerce – sometimes
in surprising and dramatic ways (Archimedes,
density and his Eureka moment).
Measurement of interactions with DataSpeaks
Interactions® can help mathematicians and
other investigators understand and model real
systems that are:
• Complex. Table 1 illustrates how DataSpeaks
Interactions® can help users investigate one
aspect of complexity – one type of treatment,
an independent variable, affecting many
dependent health variables. This one-many
type problem was addressed by dimension
reduction. DataSpeaks Interactions® also
includes tools for addressing many-one and
many-many type problems. Some of these
tools involve applying Boolean operators to
digital time series (see Table 2) for two or
more independent or dependent variables.
• Adaptive. Iterative processing as illustrated
in Figures 3 and 7 can be used to investigate
forms of adaptation such as learning,
habituation, sensitization, tolerance and
neuroplasticity. Adaptation would be
indicated by substantial changes in slope or
changes in sign.
• Stochastic. Figures 2 and 5 illustrate
how DataSpeaks’ computational algorithm
is fundamentally stochastic. Table 1
differentiates DataSpeaks’ algorithm from
classical statistics.Assessment of causality
and measurement of benefit/harm for each
individual came before statistical analysis
of group results. DataSpeaks’ algorithm,
biostatistics and quantum mechanics all use
probability. However, it often is useful to
distinguish such different disciplines.
• Nonlinear. Figure 4, parts A and B,
and Figure 6 seem to illustrate a form of
nonlinearity. Nonlinearity can be investigated
with more variables to investigate synergy and
antagonism.
• Hierarchical. DataSpeaks Interactions® can
help foster interdisciplinary investigations
by measuring interactions over time within
and between time series at different levels of
investigation such as cell, organ and organ
system and at biological, psychological and
social levels.
DataSpeaks Interactions® Software –
A Killer App for 21st Century Systems Science and P4 Medicine?
CURTIS A. BAGNE, PH.D.
DATASPEAKS, INC.
International Symposium, Institute for Systems Biology, April 2008
6 DataSpeaks Interactions® Can Help Expedite P4 Medicine
Drs. Hood and Zerhouni champion P4
medicine. DataSpeaks Interactions® can help
expedite medicine that is:
• Predictive. Measures of interaction over
time that describe past behavior can help
predict future behavior. Furthermore,
DataSpeaks’ algorithm measures time
dependent emergent system properties.
Such measures can be used to identify
genetic and other predictors, develop better
classifications of chronic health problems
and increase the value of genotyping.
• Personalized. Two keys to personalized
medicine are assessing causality for
individual patients (Table 1) and identifying
optimal minimal drug doses (Figure
6). Personalized medicine that improves
individual health will improve group average
or public health.
• Preventive or Preemptive. Twenty-first
Century science that improves scientific
understanding can be used to help prevent
health problems and control health care
costs.
• Participatory. DataSpeaks Interactions®
implemented on the Internet would help
empower people by themselves to achieve
better health for themselves and their loved
ones based on the collection and processing
of data about themselves as illustrated
in Demonstration 2. In addition, this
could help create Web-based cumulative
repositories of anonymous data and
information (as in HealthVault.com?) that
could be of great value to decision makers
such as patients, clinicians, pharmaceutical
companies, health care providers, regulators
and payers.
Is it acceptable to prescribe and administer
drugs without monitoring drug effects as
illustrated in Figure 7 when there is significant
uncertainty about safety and cost-effectiveness?
DataSpeaks Interactions® has the potential
to make electronic medical records more
attractive to clinicians and patients as they
seek better outcomes. DataSpeaks Interactions®
has potential to help create second generation
evidence-based medicine. New software
technology can help fix health care.
DataSpeaks Interactions® can raise standards
of evidence for much of systems science and
medicine.
We are at the cusp of a revolution toward 21st
Century systems science and P4 medicine.
DataSpeaks Interactions® can help advance
this revolution.
7 The Digital Promise of DataSpeaks Interactions® Software
Digitalization of biology and
medicine is a “revolution that will
transform medicine even more
than digitalization transformed
information technology and
communications”
—Dr. Leroy Hood
DataSpeaks Interactions® embodies a
computational algorithm that essentially is
digital in a manner that goes beyond the way
much other software, such as software for
classical statistics, has been made to be digital
for digital computers.
This innovative digital nature of DataSpeaks
Interactions® is illustrated by showing part of
how the DataSpeaks’ algorithm was used to
compute the benefit/harm score (8.92) with
respect to diastolic blood pressure (DBP) for
Patient 1 in Table 1. One crucial step is to
transform the analog time series for dose, an
exogenous independent action variable, and the
analog time series for DBP, a dependent action
variable, into sets of digital time series as shown
in Table 2. There are various ways to transform
analog series (a time series with more than two
levels) into a set of digital series, often without
necessary loss of information.
Table 2
Digitalization of analog time series for Dose and DBP for Patient 1
Week
1
2
3
20 20 40
Dose
DBP
≥ 20
≥ 40
= 80
≥ 78
≥ 80
≥ 81
≥ 82
≥ 84
≥ 85
≥ 87
≥ 91
≥ 93
≥ 96
≥ 98
1
0
0
1
0
0
1
1
0
4
5 6 7 8 9 10 11 12 13 14 15 16
One Analog Time Series for Dose
40 0 0 40 40 40 40 20 20 80 80 40 40
Set of Three Digital Time Series for Dose
1 0 0 1 1 1 1 1 1 1 1 1 1
1 0 0 1 1 1 1 0 0 1 1 1 1
0 0 0 0 0 0 0 0 0 1 1 0 0
One Analog Time Series for DBP
96 98 85 81 91 96 84 87 80 78 93 98 82 77 81 78
Set of Eleven Digital Time Series for DBP
1 1 1 1 1 1 1 1 1 1 1 1 1 0 1 1
1 1 1 1 1 1 1 1 1 0 1 1 1 0 1 0
1 1 1 1 1 1 1 1 0 0 1 1 1 0 1 0
1 1 1 0 1 1 1 1 0 0 1 1 1 0 0 0
1 1 1 0 1 1 1 1 0 0 1 1 0 0 0 0
1 1 1 0 1 1 0 1 0 0 1 1 0 0 0 0
1 1 0 0 1 1 0 1 0 0 1 1 0 0 0 0
1 1 0 0 1 1 0 0 0 0 1 1 0 0 0 0
1 1 0 0 0 1 0 0 0 0 1 1 0 0 0 0
1 1 0 0 0 1 0 0 0 0 0 1 0 0 0 0
0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0
DataSpeaks Interactions® Software –
A Killer App for 21st Century Systems Science and P4 Medicine?
CURTIS A. BAGNE, PH.D.
DATASPEAKS, INC.
International Symposium, Institute for Systems Biology, April 2008
Cross-classification of independent and
dependent events: Another crucial step in
DataSpeaks’ algorithm is to cross-classify each
digital series for the independent variable,
dose, with each digital series for the dependent
variable, DBP. The cell labels for this crossclassification are shown in Table 3.
to Bagne). These four cell frequencies can
help account for much complexity in how
systems work in a more digital science much
as A, T, C and G of the genetic code underlies
much complexity in how living systems
work. DataSpeaks’ computational algorithm is
basically simple.
Count the number of repeated measurement
times for each cell. Interaction scores and
benefit/harm scores are computed from the
resulting cell frequencies – a, b, c and d.
Computational details are published and
in Patent 6,317,700 (see Patents Issued
Both spatial order or sequence (snapshots
and maps for statics) and temporal order or
sequence (movies for dynamics of interaction)
are important. DataSpeaks Interactions® finds
and measures patterns of interaction over time
in temporal order.
Table 3
Digital Independent
Events
Present, 1 Absent, 0
Digital
Dependent
Events
Present, 1
Absent, 0
1,1
a=0
1,0
c=10
0,1
b=6
0,0
d=0
a+c=10
b+d=6
a+b=6
c+d=10
The distributions of potential scores shown
in Figures 2 and 5 show all scores that are
possible given the marginal frequencies of the
corresponding observed 2 x 2 tables.
Table 4: Digital pattern finding for multiple time series data.
Table 4 shows the cross-classification
portion of DataSpeak’s algorithm that
was used to obtain the summary score of
8.92. DBP was lower on all 10 weeks
when dose was 40 or more than for
any of 6 weeks when dose was less
than 40.
(The score with a value of 8.92
summarizes a 3-dimensional array with
66 (3 x 11 x 2) scores – the 3 levels of
dose and 11 levels of DPB level shown
in Table 2 and 2 levels of persistency.
Table 4 shows only 11 of these 66
standardized benefit/harm scores.)
2 x 2 Table
Week
Dose ≥ 40
1
0
2
0
3
1
4
1
5
0
6
0
7
1
8
1
≥ 78
≥ 80
≥ 81
≥ 82
≥ 84
≥ 85
≥ 87
≥ 91
Cells
≥ 93
≥ 96
= 98
1
1
1
1
1
1
1
1
b
1
1
0
1
1
1
1
1
1
1
1
b
1
1
1
1
1
1
1
1
1
0
0
c
0
0
0
1
1
1
0
0
0
0
0
c
0
0
0
1
1
1
1
1
1
1
1
b
0
0
0
1
1
1
1
1
1
1
1
b
1
1
0
1
1
1
1
1
0
0
0
c
0
0
0
1
1
1
1
1
1
1
0
c
0
0
0
DBP
9 10 11 12 13 14 15 16 Cell Counts
1 1 0 0 1 1 1 1 a b c d
1
1
0
0
0
0
0
0
c
0
0
0
1
0
0
0
0
0
0
0
c
0
0
0
1
1
1
1
1
1
1
1
b
1
0
0
1
1
1
1
1
1
1
1
b
1
1
1
1
1
1
1
0
0
0
0
c
0
0
0
0
0
0
0
0
0
0
0
c
0
0
0
1
1
1
0
0
0
0
0
c
0
0
0
1
0
0
0
0
0
0
0
c
0
0
0
Harm
Score
9
7
6
4
3
2
1
6
6
6
6
6
6
6
1
3
4
6
7
8
9
0
0
0
0
0
0
0
.77
1.36
1.86
3.24
4.15
5.35
6.87
0
0
0
0
6
5
4
2
10
10
10
10
0
1
2
4
8.92
6.82
5.06
2.37
8 Value Now
The value of DataSpeaks Interactions® is based on the value of
fundamental principles of scientific inquiry such as use of:
• Data,
• Measurement,
• Randomized experimental control and
• Operational definitions.
Demonstrations 1 and 2 show how users of DataSpeaks Interactions® can
apply these principles more extensively and with less confounding. It is
not too early to try DataSpeaks Interactions® Software.
Publication:
Bagne CA, Lewis RF. Evaluating the effects of drugs on
behavior and quality of life: an alternative strategy for
clinical trials. Journal of Consulting and Clinical Psychology
1992; 60:225-239. (http://www.dataspeaks.com/resources/
APA-JCCP-1992-Vol60-No2-P225-239.pdf)
DataSpeaks Interactions® has potential to help
establish a new high standard of empirical evidence
for much of systems science and medicine. It can
drive more extensive use of computing and the
World Wide Web for understanding people and other
types of systems. This, together with its potential
to improve health and welfare, can help make
DataSpeaks Interactions® a killer app.
Patents issued to Bagne:
6,317,700
Computational Method and System to Perform
Empirical Induction
6,516,288
Method and System to Construct Action
Coordination Profiles
Please Contact:
Curtis A. Bagne, Ph.D.
2971 Vineyards Drive
Troy, MI 48098
248 952-1968
[email protected]
www.dataspeaks.com